writeback: bdi write bandwidth estimation
[linux-2.6-block.git] / mm / backing-dev.c
CommitLineData
3fcfab16
AM
1
2#include <linux/wait.h>
3#include <linux/backing-dev.h>
03ba3782
JA
4#include <linux/kthread.h>
5#include <linux/freezer.h>
3fcfab16 6#include <linux/fs.h>
26160158 7#include <linux/pagemap.h>
03ba3782 8#include <linux/mm.h>
3fcfab16
AM
9#include <linux/sched.h>
10#include <linux/module.h>
cf0ca9fe
PZ
11#include <linux/writeback.h>
12#include <linux/device.h>
455b2864 13#include <trace/events/writeback.h>
cf0ca9fe 14
c3c53206
JA
15static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
16
26160158 17struct backing_dev_info default_backing_dev_info = {
d993831f 18 .name = "default",
26160158
JA
19 .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
20 .state = 0,
21 .capabilities = BDI_CAP_MAP_COPY,
26160158
JA
22};
23EXPORT_SYMBOL_GPL(default_backing_dev_info);
cf0ca9fe 24
5129a469
JE
25struct backing_dev_info noop_backing_dev_info = {
26 .name = "noop",
976e48f8 27 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
5129a469
JE
28};
29EXPORT_SYMBOL_GPL(noop_backing_dev_info);
30
cf0ca9fe 31static struct class *bdi_class;
cfc4ba53
JA
32
33/*
34 * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as
35 * reader side protection for bdi_pending_list. bdi_list has RCU reader side
36 * locking.
37 */
03ba3782 38DEFINE_SPINLOCK(bdi_lock);
66f3b8e2 39LIST_HEAD(bdi_list);
03ba3782
JA
40LIST_HEAD(bdi_pending_list);
41
42static struct task_struct *sync_supers_tsk;
43static struct timer_list sync_supers_timer;
44
45static int bdi_sync_supers(void *);
46static void sync_supers_timer_fn(unsigned long);
03ba3782 47
f758eeab
CH
48void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2)
49{
50 if (wb1 < wb2) {
51 spin_lock(&wb1->list_lock);
52 spin_lock_nested(&wb2->list_lock, 1);
53 } else {
54 spin_lock(&wb2->list_lock);
55 spin_lock_nested(&wb1->list_lock, 1);
56 }
57}
58
76f1418b
MS
59#ifdef CONFIG_DEBUG_FS
60#include <linux/debugfs.h>
61#include <linux/seq_file.h>
62
63static struct dentry *bdi_debug_root;
64
65static void bdi_debug_init(void)
66{
67 bdi_debug_root = debugfs_create_dir("bdi", NULL);
68}
69
70static int bdi_debug_stats_show(struct seq_file *m, void *v)
71{
72 struct backing_dev_info *bdi = m->private;
c1955ce3 73 struct bdi_writeback *wb = &bdi->wb;
364aeb28
DR
74 unsigned long background_thresh;
75 unsigned long dirty_thresh;
76 unsigned long bdi_thresh;
345227d7 77 unsigned long nr_dirty, nr_io, nr_more_io;
f09b00d3
JA
78 struct inode *inode;
79
345227d7 80 nr_dirty = nr_io = nr_more_io = 0;
f758eeab 81 spin_lock(&wb->list_lock);
7ccf19a8 82 list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
c1955ce3 83 nr_dirty++;
7ccf19a8 84 list_for_each_entry(inode, &wb->b_io, i_wb_list)
c1955ce3 85 nr_io++;
7ccf19a8 86 list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
c1955ce3 87 nr_more_io++;
f758eeab 88 spin_unlock(&wb->list_lock);
76f1418b 89
16c4042f
WF
90 global_dirty_limits(&background_thresh, &dirty_thresh);
91 bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
76f1418b
MS
92
93#define K(x) ((x) << (PAGE_SHIFT - 10))
94 seq_printf(m,
95 "BdiWriteback: %8lu kB\n"
96 "BdiReclaimable: %8lu kB\n"
97 "BdiDirtyThresh: %8lu kB\n"
98 "DirtyThresh: %8lu kB\n"
f09b00d3 99 "BackgroundThresh: %8lu kB\n"
f7d2b1ec 100 "BdiWritten: %8lu kB\n"
f09b00d3
JA
101 "b_dirty: %8lu\n"
102 "b_io: %8lu\n"
103 "b_more_io: %8lu\n"
104 "bdi_list: %8u\n"
c1955ce3 105 "state: %8lx\n",
76f1418b
MS
106 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
107 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
f7d2b1ec
JK
108 K(bdi_thresh),
109 K(dirty_thresh),
110 K(background_thresh),
111 (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
112 nr_dirty,
113 nr_io,
114 nr_more_io,
c1955ce3 115 !list_empty(&bdi->bdi_list), bdi->state);
76f1418b
MS
116#undef K
117
118 return 0;
119}
120
121static int bdi_debug_stats_open(struct inode *inode, struct file *file)
122{
123 return single_open(file, bdi_debug_stats_show, inode->i_private);
124}
125
126static const struct file_operations bdi_debug_stats_fops = {
127 .open = bdi_debug_stats_open,
128 .read = seq_read,
129 .llseek = seq_lseek,
130 .release = single_release,
131};
132
133static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
134{
135 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
136 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
137 bdi, &bdi_debug_stats_fops);
138}
139
140static void bdi_debug_unregister(struct backing_dev_info *bdi)
141{
142 debugfs_remove(bdi->debug_stats);
143 debugfs_remove(bdi->debug_dir);
144}
145#else
146static inline void bdi_debug_init(void)
147{
148}
149static inline void bdi_debug_register(struct backing_dev_info *bdi,
150 const char *name)
151{
152}
153static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
154{
155}
156#endif
157
cf0ca9fe
PZ
158static ssize_t read_ahead_kb_store(struct device *dev,
159 struct device_attribute *attr,
160 const char *buf, size_t count)
161{
162 struct backing_dev_info *bdi = dev_get_drvdata(dev);
163 char *end;
164 unsigned long read_ahead_kb;
165 ssize_t ret = -EINVAL;
166
167 read_ahead_kb = simple_strtoul(buf, &end, 10);
168 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
169 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
170 ret = count;
171 }
172 return ret;
173}
174
175#define K(pages) ((pages) << (PAGE_SHIFT - 10))
176
177#define BDI_SHOW(name, expr) \
178static ssize_t name##_show(struct device *dev, \
179 struct device_attribute *attr, char *page) \
180{ \
181 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
182 \
183 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
184}
185
186BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
187
189d3c4a
PZ
188static ssize_t min_ratio_store(struct device *dev,
189 struct device_attribute *attr, const char *buf, size_t count)
190{
191 struct backing_dev_info *bdi = dev_get_drvdata(dev);
192 char *end;
193 unsigned int ratio;
194 ssize_t ret = -EINVAL;
195
196 ratio = simple_strtoul(buf, &end, 10);
197 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
198 ret = bdi_set_min_ratio(bdi, ratio);
199 if (!ret)
200 ret = count;
201 }
202 return ret;
203}
204BDI_SHOW(min_ratio, bdi->min_ratio)
205
a42dde04
PZ
206static ssize_t max_ratio_store(struct device *dev,
207 struct device_attribute *attr, const char *buf, size_t count)
208{
209 struct backing_dev_info *bdi = dev_get_drvdata(dev);
210 char *end;
211 unsigned int ratio;
212 ssize_t ret = -EINVAL;
213
214 ratio = simple_strtoul(buf, &end, 10);
215 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
216 ret = bdi_set_max_ratio(bdi, ratio);
217 if (!ret)
218 ret = count;
219 }
220 return ret;
221}
222BDI_SHOW(max_ratio, bdi->max_ratio)
223
cf0ca9fe
PZ
224#define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
225
226static struct device_attribute bdi_dev_attrs[] = {
227 __ATTR_RW(read_ahead_kb),
189d3c4a 228 __ATTR_RW(min_ratio),
a42dde04 229 __ATTR_RW(max_ratio),
cf0ca9fe
PZ
230 __ATTR_NULL,
231};
232
233static __init int bdi_class_init(void)
234{
235 bdi_class = class_create(THIS_MODULE, "bdi");
14421453
AB
236 if (IS_ERR(bdi_class))
237 return PTR_ERR(bdi_class);
238
cf0ca9fe 239 bdi_class->dev_attrs = bdi_dev_attrs;
76f1418b 240 bdi_debug_init();
cf0ca9fe
PZ
241 return 0;
242}
76f1418b 243postcore_initcall(bdi_class_init);
cf0ca9fe 244
26160158
JA
245static int __init default_bdi_init(void)
246{
247 int err;
248
03ba3782
JA
249 sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
250 BUG_ON(IS_ERR(sync_supers_tsk));
251
03ba3782 252 setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
6423104b 253 bdi_arm_supers_timer();
03ba3782 254
26160158
JA
255 err = bdi_init(&default_backing_dev_info);
256 if (!err)
257 bdi_register(&default_backing_dev_info, NULL, "default");
976e48f8 258 err = bdi_init(&noop_backing_dev_info);
26160158
JA
259
260 return err;
261}
262subsys_initcall(default_bdi_init);
263
03ba3782
JA
264int bdi_has_dirty_io(struct backing_dev_info *bdi)
265{
266 return wb_has_dirty_io(&bdi->wb);
267}
268
03ba3782 269/*
6f904ff0 270 * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
03ba3782
JA
271 * or we risk deadlocking on ->s_umount. The longer term solution would be
272 * to implement sync_supers_bdi() or similar and simply do it from the
6f904ff0 273 * bdi writeback thread individually.
03ba3782
JA
274 */
275static int bdi_sync_supers(void *unused)
276{
277 set_user_nice(current, 0);
278
279 while (!kthread_should_stop()) {
280 set_current_state(TASK_INTERRUPTIBLE);
281 schedule();
282
283 /*
284 * Do this periodically, like kupdated() did before.
285 */
286 sync_supers();
287 }
288
289 return 0;
290}
291
6423104b 292void bdi_arm_supers_timer(void)
03ba3782
JA
293{
294 unsigned long next;
295
6423104b
JA
296 if (!dirty_writeback_interval)
297 return;
298
03ba3782
JA
299 next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
300 mod_timer(&sync_supers_timer, round_jiffies_up(next));
301}
302
303static void sync_supers_timer_fn(unsigned long unused)
304{
305 wake_up_process(sync_supers_tsk);
6423104b 306 bdi_arm_supers_timer();
03ba3782
JA
307}
308
6467716a
AB
309static void wakeup_timer_fn(unsigned long data)
310{
311 struct backing_dev_info *bdi = (struct backing_dev_info *)data;
312
313 spin_lock_bh(&bdi->wb_lock);
314 if (bdi->wb.task) {
60332023 315 trace_writeback_wake_thread(bdi);
6467716a
AB
316 wake_up_process(bdi->wb.task);
317 } else {
318 /*
319 * When bdi tasks are inactive for long time, they are killed.
320 * In this case we have to wake-up the forker thread which
321 * should create and run the bdi thread.
322 */
60332023 323 trace_writeback_wake_forker_thread(bdi);
6467716a
AB
324 wake_up_process(default_backing_dev_info.wb.task);
325 }
326 spin_unlock_bh(&bdi->wb_lock);
327}
328
329/*
330 * This function is used when the first inode for this bdi is marked dirty. It
331 * wakes-up the corresponding bdi thread which should then take care of the
332 * periodic background write-out of dirty inodes. Since the write-out would
333 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
334 * set up a timer which wakes the bdi thread up later.
335 *
336 * Note, we wouldn't bother setting up the timer, but this function is on the
337 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
338 * by delaying the wake-up.
339 */
340void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
341{
342 unsigned long timeout;
343
344 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
345 mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout);
346}
347
fff5b85a
AB
348/*
349 * Calculate the longest interval (jiffies) bdi threads are allowed to be
350 * inactive.
351 */
352static unsigned long bdi_longest_inactive(void)
353{
354 unsigned long interval;
355
356 interval = msecs_to_jiffies(dirty_writeback_interval * 10);
357 return max(5UL * 60 * HZ, interval);
358}
359
6f904ff0 360static int bdi_forker_thread(void *ptr)
03ba3782
JA
361{
362 struct bdi_writeback *me = ptr;
363
766f9164 364 current->flags |= PF_SWAPWRITE;
c1955ce3
CH
365 set_freezable();
366
367 /*
368 * Our parent may run at a different priority, just set us to normal
369 */
370 set_user_nice(current, 0);
03ba3782
JA
371
372 for (;;) {
fff5b85a 373 struct task_struct *task = NULL;
78c40cb6 374 struct backing_dev_info *bdi;
adf39240
AB
375 enum {
376 NO_ACTION, /* Nothing to do */
377 FORK_THREAD, /* Fork bdi thread */
fff5b85a 378 KILL_THREAD, /* Kill inactive bdi thread */
adf39240 379 } action = NO_ACTION;
03ba3782
JA
380
381 /*
382 * Temporary measure, we want to make sure we don't see
383 * dirty data on the default backing_dev_info
384 */
6467716a
AB
385 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) {
386 del_timer(&me->wakeup_timer);
03ba3782 387 wb_do_writeback(me, 0);
6467716a 388 }
03ba3782 389
cfc4ba53 390 spin_lock_bh(&bdi_lock);
c5f7ad23 391 set_current_state(TASK_INTERRUPTIBLE);
03ba3782 392
78c40cb6 393 list_for_each_entry(bdi, &bdi_list, bdi_list) {
adf39240
AB
394 bool have_dirty_io;
395
396 if (!bdi_cap_writeback_dirty(bdi) ||
397 bdi_cap_flush_forker(bdi))
03ba3782
JA
398 continue;
399
080dcec4
AB
400 WARN(!test_bit(BDI_registered, &bdi->state),
401 "bdi %p/%s is not registered!\n", bdi, bdi->name);
402
adf39240
AB
403 have_dirty_io = !list_empty(&bdi->work_list) ||
404 wb_has_dirty_io(&bdi->wb);
78c40cb6
AB
405
406 /*
adf39240
AB
407 * If the bdi has work to do, but the thread does not
408 * exist - create it.
78c40cb6 409 */
adf39240
AB
410 if (!bdi->wb.task && have_dirty_io) {
411 /*
412 * Set the pending bit - if someone will try to
413 * unregister this bdi - it'll wait on this bit.
414 */
415 set_bit(BDI_pending, &bdi->state);
416 action = FORK_THREAD;
417 break;
418 }
fff5b85a 419
6bf05d03
JA
420 spin_lock(&bdi->wb_lock);
421
fff5b85a
AB
422 /*
423 * If there is no work to do and the bdi thread was
424 * inactive long enough - kill it. The wb_lock is taken
425 * to make sure no-one adds more work to this bdi and
426 * wakes the bdi thread up.
427 */
428 if (bdi->wb.task && !have_dirty_io &&
429 time_after(jiffies, bdi->wb.last_active +
430 bdi_longest_inactive())) {
431 task = bdi->wb.task;
432 bdi->wb.task = NULL;
433 spin_unlock(&bdi->wb_lock);
434 set_bit(BDI_pending, &bdi->state);
435 action = KILL_THREAD;
436 break;
437 }
6bf05d03 438 spin_unlock(&bdi->wb_lock);
03ba3782 439 }
080dcec4 440 spin_unlock_bh(&bdi_lock);
03ba3782 441
c4ec7908
AB
442 /* Keep working if default bdi still has things to do */
443 if (!list_empty(&me->bdi->work_list))
444 __set_current_state(TASK_RUNNING);
445
adf39240
AB
446 switch (action) {
447 case FORK_THREAD:
448 __set_current_state(TASK_RUNNING);
6628bc74
AB
449 task = kthread_create(bdi_writeback_thread, &bdi->wb,
450 "flush-%s", dev_name(bdi->dev));
adf39240
AB
451 if (IS_ERR(task)) {
452 /*
453 * If thread creation fails, force writeout of
d46db3d5
WF
454 * the bdi from the thread. Hopefully 1024 is
455 * large enough for efficient IO.
adf39240 456 */
d46db3d5 457 writeback_inodes_wb(&bdi->wb, 1024);
fff5b85a
AB
458 } else {
459 /*
460 * The spinlock makes sure we do not lose
461 * wake-ups when racing with 'bdi_queue_work()'.
6628bc74
AB
462 * And as soon as the bdi thread is visible, we
463 * can start it.
fff5b85a 464 */
6467716a 465 spin_lock_bh(&bdi->wb_lock);
adf39240 466 bdi->wb.task = task;
6467716a 467 spin_unlock_bh(&bdi->wb_lock);
6628bc74 468 wake_up_process(task);
fff5b85a
AB
469 }
470 break;
471
472 case KILL_THREAD:
473 __set_current_state(TASK_RUNNING);
474 kthread_stop(task);
adf39240 475 break;
03ba3782 476
adf39240 477 case NO_ACTION:
253c34e9
AB
478 if (!wb_has_dirty_io(me) || !dirty_writeback_interval)
479 /*
480 * There are no dirty data. The only thing we
481 * should now care about is checking for
482 * inactive bdi threads and killing them. Thus,
483 * let's sleep for longer time, save energy and
484 * be friendly for battery-driven devices.
485 */
486 schedule_timeout(bdi_longest_inactive());
6423104b 487 else
253c34e9 488 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
03ba3782 489 try_to_freeze();
adf39240 490 /* Back to the main loop */
03ba3782
JA
491 continue;
492 }
fff5b85a
AB
493
494 /*
495 * Clear pending bit and wakeup anybody waiting to tear us down.
496 */
497 clear_bit(BDI_pending, &bdi->state);
498 smp_mb__after_clear_bit();
499 wake_up_bit(&bdi->state, BDI_pending);
03ba3782
JA
500 }
501
502 return 0;
503}
504
cfc4ba53
JA
505/*
506 * Remove bdi from bdi_list, and ensure that it is no longer visible
507 */
508static void bdi_remove_from_list(struct backing_dev_info *bdi)
509{
510 spin_lock_bh(&bdi_lock);
511 list_del_rcu(&bdi->bdi_list);
512 spin_unlock_bh(&bdi_lock);
513
514 synchronize_rcu();
515}
516
cf0ca9fe
PZ
517int bdi_register(struct backing_dev_info *bdi, struct device *parent,
518 const char *fmt, ...)
519{
cf0ca9fe 520 va_list args;
cf0ca9fe
PZ
521 struct device *dev;
522
69fc208b 523 if (bdi->dev) /* The driver needs to use separate queues per device */
c284de61 524 return 0;
f1d0b063 525
cf0ca9fe 526 va_start(args, fmt);
19051c50 527 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
cf0ca9fe 528 va_end(args);
c284de61
AB
529 if (IS_ERR(dev))
530 return PTR_ERR(dev);
66f3b8e2 531
cf0ca9fe 532 bdi->dev = dev;
cf0ca9fe 533
03ba3782
JA
534 /*
535 * Just start the forker thread for our default backing_dev_info,
536 * and add other bdi's to the list. They will get a thread created
537 * on-demand when they need it.
538 */
539 if (bdi_cap_flush_forker(bdi)) {
540 struct bdi_writeback *wb = &bdi->wb;
541
6f904ff0 542 wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s",
03ba3782 543 dev_name(dev));
c284de61
AB
544 if (IS_ERR(wb->task))
545 return PTR_ERR(wb->task);
03ba3782
JA
546 }
547
548 bdi_debug_register(bdi, dev_name(dev));
500b067c 549 set_bit(BDI_registered, &bdi->state);
c284de61
AB
550
551 spin_lock_bh(&bdi_lock);
552 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
553 spin_unlock_bh(&bdi_lock);
554
455b2864 555 trace_writeback_bdi_register(bdi);
c284de61 556 return 0;
cf0ca9fe
PZ
557}
558EXPORT_SYMBOL(bdi_register);
559
560int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
561{
562 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
563}
564EXPORT_SYMBOL(bdi_register_dev);
565
03ba3782
JA
566/*
567 * Remove bdi from the global list and shutdown any threads we have running
568 */
569static void bdi_wb_shutdown(struct backing_dev_info *bdi)
66f3b8e2 570{
03ba3782
JA
571 if (!bdi_cap_writeback_dirty(bdi))
572 return;
573
574 /*
fff5b85a 575 * Make sure nobody finds us on the bdi_list anymore
03ba3782 576 */
fff5b85a 577 bdi_remove_from_list(bdi);
03ba3782
JA
578
579 /*
fff5b85a 580 * If setup is pending, wait for that to complete first
03ba3782 581 */
fff5b85a
AB
582 wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
583 TASK_UNINTERRUPTIBLE);
03ba3782
JA
584
585 /*
c1955ce3 586 * Finally, kill the kernel thread. We don't need to be RCU
c62b17a5
RD
587 * safe anymore, since the bdi is gone from visibility. Force
588 * unfreeze of the thread before calling kthread_stop(), otherwise
589 * it would never exet if it is currently stuck in the refrigerator.
03ba3782 590 */
c1955ce3
CH
591 if (bdi->wb.task) {
592 thaw_process(bdi->wb.task);
593 kthread_stop(bdi->wb.task);
c62b17a5 594 }
66f3b8e2
JA
595}
596
592b09a4
JA
597/*
598 * This bdi is going away now, make sure that no super_blocks point to it
599 */
600static void bdi_prune_sb(struct backing_dev_info *bdi)
601{
602 struct super_block *sb;
603
604 spin_lock(&sb_lock);
605 list_for_each_entry(sb, &super_blocks, s_list) {
606 if (sb->s_bdi == bdi)
95f28604 607 sb->s_bdi = &default_backing_dev_info;
592b09a4
JA
608 }
609 spin_unlock(&sb_lock);
610}
611
cf0ca9fe
PZ
612void bdi_unregister(struct backing_dev_info *bdi)
613{
614 if (bdi->dev) {
455b2864 615 trace_writeback_bdi_unregister(bdi);
8c4db335 616 bdi_prune_sb(bdi);
6467716a 617 del_timer_sync(&bdi->wb.wakeup_timer);
8c4db335 618
03ba3782
JA
619 if (!bdi_cap_flush_forker(bdi))
620 bdi_wb_shutdown(bdi);
76f1418b 621 bdi_debug_unregister(bdi);
cf0ca9fe
PZ
622 device_unregister(bdi->dev);
623 bdi->dev = NULL;
624 }
625}
626EXPORT_SYMBOL(bdi_unregister);
3fcfab16 627
6467716a
AB
628static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
629{
630 memset(wb, 0, sizeof(*wb));
631
632 wb->bdi = bdi;
633 wb->last_old_flush = jiffies;
634 INIT_LIST_HEAD(&wb->b_dirty);
635 INIT_LIST_HEAD(&wb->b_io);
636 INIT_LIST_HEAD(&wb->b_more_io);
f758eeab 637 spin_lock_init(&wb->list_lock);
6467716a
AB
638 setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi);
639}
640
e98be2d5
WF
641/*
642 * Initial write bandwidth: 100 MB/s
643 */
644#define INIT_BW (100 << (20 - PAGE_SHIFT))
645
b2e8fb6e
PZ
646int bdi_init(struct backing_dev_info *bdi)
647{
03ba3782 648 int i, err;
b2e8fb6e 649
cf0ca9fe
PZ
650 bdi->dev = NULL;
651
189d3c4a 652 bdi->min_ratio = 0;
a42dde04
PZ
653 bdi->max_ratio = 100;
654 bdi->max_prop_frac = PROP_FRAC_BASE;
03ba3782 655 spin_lock_init(&bdi->wb_lock);
66f3b8e2 656 INIT_LIST_HEAD(&bdi->bdi_list);
03ba3782
JA
657 INIT_LIST_HEAD(&bdi->work_list);
658
659 bdi_wb_init(&bdi->wb, bdi);
660
b2e8fb6e 661 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
ea319518 662 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
04fbfdc1
PZ
663 if (err)
664 goto err;
665 }
666
667 bdi->dirty_exceeded = 0;
e98be2d5
WF
668
669 bdi->bw_time_stamp = jiffies;
670 bdi->written_stamp = 0;
671
672 bdi->write_bandwidth = INIT_BW;
673 bdi->avg_write_bandwidth = INIT_BW;
674
04fbfdc1
PZ
675 err = prop_local_init_percpu(&bdi->completions);
676
677 if (err) {
678err:
4b01a0b1 679 while (i--)
04fbfdc1 680 percpu_counter_destroy(&bdi->bdi_stat[i]);
b2e8fb6e
PZ
681 }
682
683 return err;
684}
685EXPORT_SYMBOL(bdi_init);
686
687void bdi_destroy(struct backing_dev_info *bdi)
688{
689 int i;
690
ce5f8e77
JA
691 /*
692 * Splice our entries to the default_backing_dev_info, if this
693 * bdi disappears
694 */
695 if (bdi_has_dirty_io(bdi)) {
696 struct bdi_writeback *dst = &default_backing_dev_info.wb;
697
f758eeab 698 bdi_lock_two(&bdi->wb, dst);
ce5f8e77
JA
699 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
700 list_splice(&bdi->wb.b_io, &dst->b_io);
701 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
f758eeab
CH
702 spin_unlock(&bdi->wb.list_lock);
703 spin_unlock(&dst->list_lock);
ce5f8e77 704 }
66f3b8e2 705
cf0ca9fe
PZ
706 bdi_unregister(bdi);
707
b2e8fb6e
PZ
708 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
709 percpu_counter_destroy(&bdi->bdi_stat[i]);
04fbfdc1
PZ
710
711 prop_local_destroy_percpu(&bdi->completions);
b2e8fb6e
PZ
712}
713EXPORT_SYMBOL(bdi_destroy);
714
c3c53206
JA
715/*
716 * For use from filesystems to quickly init and register a bdi associated
717 * with dirty writeback
718 */
719int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
720 unsigned int cap)
721{
722 char tmp[32];
723 int err;
724
725 bdi->name = name;
726 bdi->capabilities = cap;
727 err = bdi_init(bdi);
728 if (err)
729 return err;
730
731 sprintf(tmp, "%.28s%s", name, "-%d");
732 err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
733 if (err) {
734 bdi_destroy(bdi);
735 return err;
736 }
737
738 return 0;
739}
740EXPORT_SYMBOL(bdi_setup_and_register);
741
3fcfab16
AM
742static wait_queue_head_t congestion_wqh[2] = {
743 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
744 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
745 };
0e093d99 746static atomic_t nr_bdi_congested[2];
3fcfab16 747
1faa16d2 748void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
3fcfab16
AM
749{
750 enum bdi_state bit;
1faa16d2 751 wait_queue_head_t *wqh = &congestion_wqh[sync];
3fcfab16 752
1faa16d2 753 bit = sync ? BDI_sync_congested : BDI_async_congested;
0e093d99
MG
754 if (test_and_clear_bit(bit, &bdi->state))
755 atomic_dec(&nr_bdi_congested[sync]);
3fcfab16
AM
756 smp_mb__after_clear_bit();
757 if (waitqueue_active(wqh))
758 wake_up(wqh);
759}
760EXPORT_SYMBOL(clear_bdi_congested);
761
1faa16d2 762void set_bdi_congested(struct backing_dev_info *bdi, int sync)
3fcfab16
AM
763{
764 enum bdi_state bit;
765
1faa16d2 766 bit = sync ? BDI_sync_congested : BDI_async_congested;
0e093d99
MG
767 if (!test_and_set_bit(bit, &bdi->state))
768 atomic_inc(&nr_bdi_congested[sync]);
3fcfab16
AM
769}
770EXPORT_SYMBOL(set_bdi_congested);
771
772/**
773 * congestion_wait - wait for a backing_dev to become uncongested
8aa7e847 774 * @sync: SYNC or ASYNC IO
3fcfab16
AM
775 * @timeout: timeout in jiffies
776 *
777 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
778 * write congestion. If no backing_devs are congested then just wait for the
779 * next write to be completed.
780 */
8aa7e847 781long congestion_wait(int sync, long timeout)
3fcfab16
AM
782{
783 long ret;
52bb9198 784 unsigned long start = jiffies;
3fcfab16 785 DEFINE_WAIT(wait);
8aa7e847 786 wait_queue_head_t *wqh = &congestion_wqh[sync];
3fcfab16
AM
787
788 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
789 ret = io_schedule_timeout(timeout);
790 finish_wait(wqh, &wait);
52bb9198
MG
791
792 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
793 jiffies_to_usecs(jiffies - start));
794
3fcfab16
AM
795 return ret;
796}
797EXPORT_SYMBOL(congestion_wait);
04fbfdc1 798
0e093d99
MG
799/**
800 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
801 * @zone: A zone to check if it is heavily congested
802 * @sync: SYNC or ASYNC IO
803 * @timeout: timeout in jiffies
804 *
805 * In the event of a congested backing_dev (any backing_dev) and the given
806 * @zone has experienced recent congestion, this waits for up to @timeout
807 * jiffies for either a BDI to exit congestion of the given @sync queue
808 * or a write to complete.
809 *
25985edc 810 * In the absence of zone congestion, cond_resched() is called to yield
0e093d99
MG
811 * the processor if necessary but otherwise does not sleep.
812 *
813 * The return value is 0 if the sleep is for the full timeout. Otherwise,
814 * it is the number of jiffies that were still remaining when the function
815 * returned. return_value == timeout implies the function did not sleep.
816 */
817long wait_iff_congested(struct zone *zone, int sync, long timeout)
818{
819 long ret;
820 unsigned long start = jiffies;
821 DEFINE_WAIT(wait);
822 wait_queue_head_t *wqh = &congestion_wqh[sync];
823
824 /*
825 * If there is no congestion, or heavy congestion is not being
826 * encountered in the current zone, yield if necessary instead
827 * of sleeping on the congestion queue
828 */
829 if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
830 !zone_is_reclaim_congested(zone)) {
831 cond_resched();
832
833 /* In case we scheduled, work out time remaining */
834 ret = timeout - (jiffies - start);
835 if (ret < 0)
836 ret = 0;
837
838 goto out;
839 }
840
841 /* Sleep until uncongested or a write happens */
842 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
843 ret = io_schedule_timeout(timeout);
844 finish_wait(wqh, &wait);
845
846out:
847 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
848 jiffies_to_usecs(jiffies - start));
849
850 return ret;
851}
852EXPORT_SYMBOL(wait_iff_congested);